Troubleshooting Analyses of Production Data

Production data present substantial challenges for statistical analysis, but they may hold information of great value in resolving persistent manufacturing deficiencies. Because these data are not generated in controlled experiments, frequently key factors are confounded or nested, observations are unbalanced across factor levels, and a substantial number of values may be missing. On the other hand, these data represent the most realistic characterization of the manufacturing process on a production scale. This paper aims to describe and illustrate, with a case study adapted from an industrial project, how production data can be analyzed to provide clues to the sources of quality problems. Although such observational studies cannot definitively prove the existence of a cause-and-effect mechanism, results of troubleshooting analyses may suggest potential targets for corrective actions, as well as off-line experiments or further measurements and analyses to confirm the root cause of the manufacturing problem. These investigations may be regarded as part of the analysis step in a six sigma DMAIC methodology. They are undertaken to improve the process mean with respect to specification limits and to control process variation, and their effectiveness can be measured in subsequent capability studies.